1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/8.1/sys/netinet/tcp_subr.c 207695 2010-05-06 06:44:19Z bz $");
34
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/jail.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58
59 #include <vm/uma.h>
60
61 #include <net/route.h>
62 #include <net/if.h>
63 #include <net/vnet.h>
64
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 #include <netinet/in_pcb.h>
72 #ifdef INET6
73 #include <netinet6/in6_pcb.h>
74 #endif
75 #include <netinet/in_var.h>
76 #include <netinet/ip_var.h>
77 #ifdef INET6
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82 #include <netinet/ip_icmp.h>
83 #include <netinet/tcp.h>
84 #include <netinet/tcp_fsm.h>
85 #include <netinet/tcp_seq.h>
86 #include <netinet/tcp_timer.h>
87 #include <netinet/tcp_var.h>
88 #include <netinet/tcp_syncache.h>
89 #include <netinet/tcp_offload.h>
90 #ifdef INET6
91 #include <netinet6/tcp6_var.h>
92 #endif
93 #include <netinet/tcpip.h>
94 #ifdef TCPDEBUG
95 #include <netinet/tcp_debug.h>
96 #endif
97 #include <netinet6/ip6protosw.h>
98
99 #ifdef IPSEC
100 #include <netipsec/ipsec.h>
101 #include <netipsec/xform.h>
102 #ifdef INET6
103 #include <netipsec/ipsec6.h>
104 #endif
105 #include <netipsec/key.h>
106 #include <sys/syslog.h>
107 #endif /*IPSEC*/
108
109 #include <machine/in_cksum.h>
110 #include <sys/md5.h>
111
112 #include <security/mac/mac_framework.h>
113
114 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
115 #ifdef INET6
116 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
117 #endif
118
119 static int
120 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
121 {
122 int error, new;
123
124 new = V_tcp_mssdflt;
125 error = sysctl_handle_int(oidp, &new, 0, req);
126 if (error == 0 && req->newptr) {
127 if (new < TCP_MINMSS)
128 error = EINVAL;
129 else
130 V_tcp_mssdflt = new;
131 }
132 return (error);
133 }
134
135 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
136 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
137 &sysctl_net_inet_tcp_mss_check, "I",
138 "Default TCP Maximum Segment Size");
139
140 #ifdef INET6
141 static int
142 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
143 {
144 int error, new;
145
146 new = V_tcp_v6mssdflt;
147 error = sysctl_handle_int(oidp, &new, 0, req);
148 if (error == 0 && req->newptr) {
149 if (new < TCP_MINMSS)
150 error = EINVAL;
151 else
152 V_tcp_v6mssdflt = new;
153 }
154 return (error);
155 }
156
157 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
158 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
159 &sysctl_net_inet_tcp_mss_v6_check, "I",
160 "Default TCP Maximum Segment Size for IPv6");
161 #endif
162
163 static int
164 vnet_sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
165 {
166
167 VNET_SYSCTL_ARG(req, arg1);
168 return (sysctl_msec_to_ticks(oidp, arg1, arg2, req));
169 }
170
171 /*
172 * Minimum MSS we accept and use. This prevents DoS attacks where
173 * we are forced to a ridiculous low MSS like 20 and send hundreds
174 * of packets instead of one. The effect scales with the available
175 * bandwidth and quickly saturates the CPU and network interface
176 * with packet generation and sending. Set to zero to disable MINMSS
177 * checking. This setting prevents us from sending too small packets.
178 */
179 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
180 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
181 &VNET_NAME(tcp_minmss), 0,
182 "Minmum TCP Maximum Segment Size");
183
184 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
185 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
186 &VNET_NAME(tcp_do_rfc1323), 0,
187 "Enable rfc1323 (high performance TCP) extensions");
188
189 static int tcp_log_debug = 0;
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
191 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
192
193 static int tcp_tcbhashsize = 0;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
195 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
196
197 static int do_tcpdrain = 1;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
199 "Enable tcp_drain routine for extra help when low on mbufs");
200
201 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
202 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
203
204 static VNET_DEFINE(int, icmp_may_rst) = 1;
205 #define V_icmp_may_rst VNET(icmp_may_rst)
206 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
207 &VNET_NAME(icmp_may_rst), 0,
208 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
209
210 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
211 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
212 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
213 &VNET_NAME(tcp_isn_reseed_interval), 0,
214 "Seconds between reseeding of ISN secret");
215
216 /*
217 * TCP bandwidth limiting sysctls. Note that the default lower bound of
218 * 1024 exists only for debugging. A good production default would be
219 * something like 6100.
220 */
221 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
222 "TCP inflight data limiting");
223
224 static VNET_DEFINE(int, tcp_inflight_enable) = 1;
225 #define V_tcp_inflight_enable VNET(tcp_inflight_enable)
226 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
227 &VNET_NAME(tcp_inflight_enable), 0,
228 "Enable automatic TCP inflight data limiting");
229
230 static int tcp_inflight_debug = 0;
231 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
232 &tcp_inflight_debug, 0,
233 "Debug TCP inflight calculations");
234
235 static VNET_DEFINE(int, tcp_inflight_rttthresh);
236 #define V_tcp_inflight_rttthresh VNET(tcp_inflight_rttthresh)
237 SYSCTL_VNET_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh,
238 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_inflight_rttthresh), 0,
239 vnet_sysctl_msec_to_ticks, "I",
240 "RTT threshold below which inflight will deactivate itself");
241
242 static VNET_DEFINE(int, tcp_inflight_min) = 6144;
243 #define V_tcp_inflight_min VNET(tcp_inflight_min)
244 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
245 &VNET_NAME(tcp_inflight_min), 0,
246 "Lower-bound for TCP inflight window");
247
248 static VNET_DEFINE(int, tcp_inflight_max) = TCP_MAXWIN << TCP_MAX_WINSHIFT;
249 #define V_tcp_inflight_max VNET(tcp_inflight_max)
250 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
251 &VNET_NAME(tcp_inflight_max), 0,
252 "Upper-bound for TCP inflight window");
253
254 static VNET_DEFINE(int, tcp_inflight_stab) = 20;
255 #define V_tcp_inflight_stab VNET(tcp_inflight_stab)
256 SYSCTL_VNET_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
257 &VNET_NAME(tcp_inflight_stab), 0,
258 "Inflight Algorithm Stabilization 20 = 2 packets");
259
260 VNET_DEFINE(uma_zone_t, sack_hole_zone);
261 #define V_sack_hole_zone VNET(sack_hole_zone)
262
263 static struct inpcb *tcp_notify(struct inpcb *, int);
264 static void tcp_isn_tick(void *);
265
266 /*
267 * Target size of TCP PCB hash tables. Must be a power of two.
268 *
269 * Note that this can be overridden by the kernel environment
270 * variable net.inet.tcp.tcbhashsize
271 */
272 #ifndef TCBHASHSIZE
273 #define TCBHASHSIZE 512
274 #endif
275
276 /*
277 * XXX
278 * Callouts should be moved into struct tcp directly. They are currently
279 * separate because the tcpcb structure is exported to userland for sysctl
280 * parsing purposes, which do not know about callouts.
281 */
282 struct tcpcb_mem {
283 struct tcpcb tcb;
284 struct tcp_timer tt;
285 };
286
287 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
288 #define V_tcpcb_zone VNET(tcpcb_zone)
289
290 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
291 struct callout isn_callout;
292 static struct mtx isn_mtx;
293
294 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
295 #define ISN_LOCK() mtx_lock(&isn_mtx)
296 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
297
298 /*
299 * TCP initialization.
300 */
301 static void
302 tcp_zone_change(void *tag)
303 {
304
305 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
306 uma_zone_set_max(V_tcpcb_zone, maxsockets);
307 tcp_tw_zone_change();
308 }
309
310 static int
311 tcp_inpcb_init(void *mem, int size, int flags)
312 {
313 struct inpcb *inp = mem;
314
315 INP_LOCK_INIT(inp, "inp", "tcpinp");
316 return (0);
317 }
318
319 void
320 tcp_init(void)
321 {
322 int hashsize;
323
324 INP_INFO_LOCK_INIT(&V_tcbinfo, "tcp");
325 LIST_INIT(&V_tcb);
326 #ifdef VIMAGE
327 V_tcbinfo.ipi_vnet = curvnet;
328 #endif
329 V_tcbinfo.ipi_listhead = &V_tcb;
330 hashsize = TCBHASHSIZE;
331 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
332 if (!powerof2(hashsize)) {
333 printf("WARNING: TCB hash size not a power of 2\n");
334 hashsize = 512; /* safe default */
335 }
336 V_tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
337 &V_tcbinfo.ipi_hashmask);
338 V_tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
339 &V_tcbinfo.ipi_porthashmask);
340 V_tcbinfo.ipi_zone = uma_zcreate("tcp_inpcb", sizeof(struct inpcb),
341 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
342 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
343 V_tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
344
345 /*
346 * These have to be type stable for the benefit of the timers.
347 */
348 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
349 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
350 uma_zone_set_max(V_tcpcb_zone, maxsockets);
351
352 tcp_tw_init();
353 syncache_init();
354 tcp_hc_init();
355 tcp_reass_init();
356
357 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
358 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
359 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
360
361 /* Skip initialization of globals for non-default instances. */
362 if (!IS_DEFAULT_VNET(curvnet))
363 return;
364
365 /* XXX virtualize those bellow? */
366 tcp_delacktime = TCPTV_DELACK;
367 tcp_keepinit = TCPTV_KEEP_INIT;
368 tcp_keepidle = TCPTV_KEEP_IDLE;
369 tcp_keepintvl = TCPTV_KEEPINTVL;
370 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
371 tcp_msl = TCPTV_MSL;
372 tcp_rexmit_min = TCPTV_MIN;
373 if (tcp_rexmit_min < 1)
374 tcp_rexmit_min = 1;
375 tcp_rexmit_slop = TCPTV_CPU_VAR;
376 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
377 tcp_tcbhashsize = hashsize;
378
379 #ifdef INET6
380 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
381 #else /* INET6 */
382 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
383 #endif /* INET6 */
384 if (max_protohdr < TCP_MINPROTOHDR)
385 max_protohdr = TCP_MINPROTOHDR;
386 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
387 panic("tcp_init");
388 #undef TCP_MINPROTOHDR
389
390 ISN_LOCK_INIT();
391 callout_init(&isn_callout, CALLOUT_MPSAFE);
392 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
393 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
394 SHUTDOWN_PRI_DEFAULT);
395 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
396 EVENTHANDLER_PRI_ANY);
397 }
398
399 #ifdef VIMAGE
400 void
401 tcp_destroy(void)
402 {
403
404 tcp_reass_destroy();
405 tcp_hc_destroy();
406 syncache_destroy();
407 tcp_tw_destroy();
408
409 /* XXX check that hashes are empty! */
410 hashdestroy(V_tcbinfo.ipi_hashbase, M_PCB,
411 V_tcbinfo.ipi_hashmask);
412 hashdestroy(V_tcbinfo.ipi_porthashbase, M_PCB,
413 V_tcbinfo.ipi_porthashmask);
414
415 uma_zdestroy(V_sack_hole_zone);
416 uma_zdestroy(V_tcpcb_zone);
417 uma_zdestroy(V_tcbinfo.ipi_zone);
418
419 INP_INFO_LOCK_DESTROY(&V_tcbinfo);
420 }
421 #endif
422
423 void
424 tcp_fini(void *xtp)
425 {
426
427 callout_stop(&isn_callout);
428 }
429
430 /*
431 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
432 * tcp_template used to store this data in mbufs, but we now recopy it out
433 * of the tcpcb each time to conserve mbufs.
434 */
435 void
436 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
437 {
438 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
439
440 INP_WLOCK_ASSERT(inp);
441
442 #ifdef INET6
443 if ((inp->inp_vflag & INP_IPV6) != 0) {
444 struct ip6_hdr *ip6;
445
446 ip6 = (struct ip6_hdr *)ip_ptr;
447 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
448 (inp->inp_flow & IPV6_FLOWINFO_MASK);
449 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
450 (IPV6_VERSION & IPV6_VERSION_MASK);
451 ip6->ip6_nxt = IPPROTO_TCP;
452 ip6->ip6_plen = htons(sizeof(struct tcphdr));
453 ip6->ip6_src = inp->in6p_laddr;
454 ip6->ip6_dst = inp->in6p_faddr;
455 } else
456 #endif
457 {
458 struct ip *ip;
459
460 ip = (struct ip *)ip_ptr;
461 ip->ip_v = IPVERSION;
462 ip->ip_hl = 5;
463 ip->ip_tos = inp->inp_ip_tos;
464 ip->ip_len = 0;
465 ip->ip_id = 0;
466 ip->ip_off = 0;
467 ip->ip_ttl = inp->inp_ip_ttl;
468 ip->ip_sum = 0;
469 ip->ip_p = IPPROTO_TCP;
470 ip->ip_src = inp->inp_laddr;
471 ip->ip_dst = inp->inp_faddr;
472 }
473 th->th_sport = inp->inp_lport;
474 th->th_dport = inp->inp_fport;
475 th->th_seq = 0;
476 th->th_ack = 0;
477 th->th_x2 = 0;
478 th->th_off = 5;
479 th->th_flags = 0;
480 th->th_win = 0;
481 th->th_urp = 0;
482 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
483 }
484
485 /*
486 * Create template to be used to send tcp packets on a connection.
487 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
488 * use for this function is in keepalives, which use tcp_respond.
489 */
490 struct tcptemp *
491 tcpip_maketemplate(struct inpcb *inp)
492 {
493 struct tcptemp *t;
494
495 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
496 if (t == NULL)
497 return (NULL);
498 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
499 return (t);
500 }
501
502 /*
503 * Send a single message to the TCP at address specified by
504 * the given TCP/IP header. If m == NULL, then we make a copy
505 * of the tcpiphdr at ti and send directly to the addressed host.
506 * This is used to force keep alive messages out using the TCP
507 * template for a connection. If flags are given then we send
508 * a message back to the TCP which originated the * segment ti,
509 * and discard the mbuf containing it and any other attached mbufs.
510 *
511 * In any case the ack and sequence number of the transmitted
512 * segment are as specified by the parameters.
513 *
514 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
515 */
516 void
517 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
518 tcp_seq ack, tcp_seq seq, int flags)
519 {
520 int tlen;
521 int win = 0;
522 struct ip *ip;
523 struct tcphdr *nth;
524 #ifdef INET6
525 struct ip6_hdr *ip6;
526 int isipv6;
527 #endif /* INET6 */
528 int ipflags = 0;
529 struct inpcb *inp;
530
531 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
532
533 #ifdef INET6
534 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
535 ip6 = ipgen;
536 #endif /* INET6 */
537 ip = ipgen;
538
539 if (tp != NULL) {
540 inp = tp->t_inpcb;
541 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
542 INP_WLOCK_ASSERT(inp);
543 } else
544 inp = NULL;
545
546 if (tp != NULL) {
547 if (!(flags & TH_RST)) {
548 win = sbspace(&inp->inp_socket->so_rcv);
549 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
550 win = (long)TCP_MAXWIN << tp->rcv_scale;
551 }
552 }
553 if (m == NULL) {
554 m = m_gethdr(M_DONTWAIT, MT_DATA);
555 if (m == NULL)
556 return;
557 tlen = 0;
558 m->m_data += max_linkhdr;
559 #ifdef INET6
560 if (isipv6) {
561 bcopy((caddr_t)ip6, mtod(m, caddr_t),
562 sizeof(struct ip6_hdr));
563 ip6 = mtod(m, struct ip6_hdr *);
564 nth = (struct tcphdr *)(ip6 + 1);
565 } else
566 #endif /* INET6 */
567 {
568 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
569 ip = mtod(m, struct ip *);
570 nth = (struct tcphdr *)(ip + 1);
571 }
572 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
573 flags = TH_ACK;
574 } else {
575 /*
576 * reuse the mbuf.
577 * XXX MRT We inherrit the FIB, which is lucky.
578 */
579 m_freem(m->m_next);
580 m->m_next = NULL;
581 m->m_data = (caddr_t)ipgen;
582 /* m_len is set later */
583 tlen = 0;
584 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
585 #ifdef INET6
586 if (isipv6) {
587 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
588 nth = (struct tcphdr *)(ip6 + 1);
589 } else
590 #endif /* INET6 */
591 {
592 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
593 nth = (struct tcphdr *)(ip + 1);
594 }
595 if (th != nth) {
596 /*
597 * this is usually a case when an extension header
598 * exists between the IPv6 header and the
599 * TCP header.
600 */
601 nth->th_sport = th->th_sport;
602 nth->th_dport = th->th_dport;
603 }
604 xchg(nth->th_dport, nth->th_sport, uint16_t);
605 #undef xchg
606 }
607 #ifdef INET6
608 if (isipv6) {
609 ip6->ip6_flow = 0;
610 ip6->ip6_vfc = IPV6_VERSION;
611 ip6->ip6_nxt = IPPROTO_TCP;
612 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
613 tlen));
614 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
615 } else
616 #endif
617 {
618 tlen += sizeof (struct tcpiphdr);
619 ip->ip_len = tlen;
620 ip->ip_ttl = V_ip_defttl;
621 if (V_path_mtu_discovery)
622 ip->ip_off |= IP_DF;
623 }
624 m->m_len = tlen;
625 m->m_pkthdr.len = tlen;
626 m->m_pkthdr.rcvif = NULL;
627 #ifdef MAC
628 if (inp != NULL) {
629 /*
630 * Packet is associated with a socket, so allow the
631 * label of the response to reflect the socket label.
632 */
633 INP_WLOCK_ASSERT(inp);
634 mac_inpcb_create_mbuf(inp, m);
635 } else {
636 /*
637 * Packet is not associated with a socket, so possibly
638 * update the label in place.
639 */
640 mac_netinet_tcp_reply(m);
641 }
642 #endif
643 nth->th_seq = htonl(seq);
644 nth->th_ack = htonl(ack);
645 nth->th_x2 = 0;
646 nth->th_off = sizeof (struct tcphdr) >> 2;
647 nth->th_flags = flags;
648 if (tp != NULL)
649 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
650 else
651 nth->th_win = htons((u_short)win);
652 nth->th_urp = 0;
653 #ifdef INET6
654 if (isipv6) {
655 nth->th_sum = 0;
656 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
657 sizeof(struct ip6_hdr),
658 tlen - sizeof(struct ip6_hdr));
659 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
660 NULL, NULL);
661 } else
662 #endif /* INET6 */
663 {
664 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
665 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
666 m->m_pkthdr.csum_flags = CSUM_TCP;
667 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
668 }
669 #ifdef TCPDEBUG
670 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
671 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
672 #endif
673 #ifdef INET6
674 if (isipv6)
675 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
676 else
677 #endif /* INET6 */
678 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
679 }
680
681 /*
682 * Create a new TCP control block, making an
683 * empty reassembly queue and hooking it to the argument
684 * protocol control block. The `inp' parameter must have
685 * come from the zone allocator set up in tcp_init().
686 */
687 struct tcpcb *
688 tcp_newtcpcb(struct inpcb *inp)
689 {
690 struct tcpcb_mem *tm;
691 struct tcpcb *tp;
692 #ifdef INET6
693 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
694 #endif /* INET6 */
695
696 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
697 if (tm == NULL)
698 return (NULL);
699 tp = &tm->tcb;
700 #ifdef VIMAGE
701 tp->t_vnet = inp->inp_vnet;
702 #endif
703 tp->t_timers = &tm->tt;
704 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
705 tp->t_maxseg = tp->t_maxopd =
706 #ifdef INET6
707 isipv6 ? V_tcp_v6mssdflt :
708 #endif /* INET6 */
709 V_tcp_mssdflt;
710
711 /* Set up our timeouts. */
712 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
713 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
714 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
715 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
716 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
717
718 if (V_tcp_do_rfc1323)
719 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
720 if (V_tcp_do_sack)
721 tp->t_flags |= TF_SACK_PERMIT;
722 TAILQ_INIT(&tp->snd_holes);
723 tp->t_inpcb = inp; /* XXX */
724 /*
725 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
726 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
727 * reasonable initial retransmit time.
728 */
729 tp->t_srtt = TCPTV_SRTTBASE;
730 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
731 tp->t_rttmin = tcp_rexmit_min;
732 tp->t_rxtcur = TCPTV_RTOBASE;
733 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
734 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
735 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
736 tp->t_rcvtime = ticks;
737 tp->t_bw_rtttime = ticks;
738 /*
739 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
740 * because the socket may be bound to an IPv6 wildcard address,
741 * which may match an IPv4-mapped IPv6 address.
742 */
743 inp->inp_ip_ttl = V_ip_defttl;
744 inp->inp_ppcb = tp;
745 return (tp); /* XXX */
746 }
747
748 /*
749 * Drop a TCP connection, reporting
750 * the specified error. If connection is synchronized,
751 * then send a RST to peer.
752 */
753 struct tcpcb *
754 tcp_drop(struct tcpcb *tp, int errno)
755 {
756 struct socket *so = tp->t_inpcb->inp_socket;
757
758 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
759 INP_WLOCK_ASSERT(tp->t_inpcb);
760
761 if (TCPS_HAVERCVDSYN(tp->t_state)) {
762 tp->t_state = TCPS_CLOSED;
763 (void) tcp_output_reset(tp);
764 TCPSTAT_INC(tcps_drops);
765 } else
766 TCPSTAT_INC(tcps_conndrops);
767 if (errno == ETIMEDOUT && tp->t_softerror)
768 errno = tp->t_softerror;
769 so->so_error = errno;
770 return (tcp_close(tp));
771 }
772
773 void
774 tcp_discardcb(struct tcpcb *tp)
775 {
776 struct tseg_qent *q;
777 struct inpcb *inp = tp->t_inpcb;
778 struct socket *so = inp->inp_socket;
779 #ifdef INET6
780 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
781 #endif /* INET6 */
782
783 INP_WLOCK_ASSERT(inp);
784
785 /*
786 * Make sure that all of our timers are stopped before we
787 * delete the PCB.
788 */
789 callout_stop(&tp->t_timers->tt_rexmt);
790 callout_stop(&tp->t_timers->tt_persist);
791 callout_stop(&tp->t_timers->tt_keep);
792 callout_stop(&tp->t_timers->tt_2msl);
793 callout_stop(&tp->t_timers->tt_delack);
794
795 /*
796 * If we got enough samples through the srtt filter,
797 * save the rtt and rttvar in the routing entry.
798 * 'Enough' is arbitrarily defined as 4 rtt samples.
799 * 4 samples is enough for the srtt filter to converge
800 * to within enough % of the correct value; fewer samples
801 * and we could save a bogus rtt. The danger is not high
802 * as tcp quickly recovers from everything.
803 * XXX: Works very well but needs some more statistics!
804 */
805 if (tp->t_rttupdated >= 4) {
806 struct hc_metrics_lite metrics;
807 u_long ssthresh;
808
809 bzero(&metrics, sizeof(metrics));
810 /*
811 * Update the ssthresh always when the conditions below
812 * are satisfied. This gives us better new start value
813 * for the congestion avoidance for new connections.
814 * ssthresh is only set if packet loss occured on a session.
815 *
816 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
817 * being torn down. Ideally this code would not use 'so'.
818 */
819 ssthresh = tp->snd_ssthresh;
820 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
821 /*
822 * convert the limit from user data bytes to
823 * packets then to packet data bytes.
824 */
825 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
826 if (ssthresh < 2)
827 ssthresh = 2;
828 ssthresh *= (u_long)(tp->t_maxseg +
829 #ifdef INET6
830 (isipv6 ? sizeof (struct ip6_hdr) +
831 sizeof (struct tcphdr) :
832 #endif
833 sizeof (struct tcpiphdr)
834 #ifdef INET6
835 )
836 #endif
837 );
838 } else
839 ssthresh = 0;
840 metrics.rmx_ssthresh = ssthresh;
841
842 metrics.rmx_rtt = tp->t_srtt;
843 metrics.rmx_rttvar = tp->t_rttvar;
844 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
845 metrics.rmx_bandwidth = tp->snd_bandwidth;
846 metrics.rmx_cwnd = tp->snd_cwnd;
847 metrics.rmx_sendpipe = 0;
848 metrics.rmx_recvpipe = 0;
849
850 tcp_hc_update(&inp->inp_inc, &metrics);
851 }
852
853 /* free the reassembly queue, if any */
854 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
855 LIST_REMOVE(q, tqe_q);
856 m_freem(q->tqe_m);
857 uma_zfree(V_tcp_reass_zone, q);
858 tp->t_segqlen--;
859 V_tcp_reass_qsize--;
860 }
861 /* Disconnect offload device, if any. */
862 tcp_offload_detach(tp);
863
864 tcp_free_sackholes(tp);
865 inp->inp_ppcb = NULL;
866 tp->t_inpcb = NULL;
867 uma_zfree(V_tcpcb_zone, tp);
868 }
869
870 /*
871 * Attempt to close a TCP control block, marking it as dropped, and freeing
872 * the socket if we hold the only reference.
873 */
874 struct tcpcb *
875 tcp_close(struct tcpcb *tp)
876 {
877 struct inpcb *inp = tp->t_inpcb;
878 struct socket *so;
879
880 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
881 INP_WLOCK_ASSERT(inp);
882
883 /* Notify any offload devices of listener close */
884 if (tp->t_state == TCPS_LISTEN)
885 tcp_offload_listen_close(tp);
886 in_pcbdrop(inp);
887 TCPSTAT_INC(tcps_closed);
888 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
889 so = inp->inp_socket;
890 soisdisconnected(so);
891 if (inp->inp_flags & INP_SOCKREF) {
892 KASSERT(so->so_state & SS_PROTOREF,
893 ("tcp_close: !SS_PROTOREF"));
894 inp->inp_flags &= ~INP_SOCKREF;
895 INP_WUNLOCK(inp);
896 ACCEPT_LOCK();
897 SOCK_LOCK(so);
898 so->so_state &= ~SS_PROTOREF;
899 sofree(so);
900 return (NULL);
901 }
902 return (tp);
903 }
904
905 void
906 tcp_drain(void)
907 {
908 VNET_ITERATOR_DECL(vnet_iter);
909
910 if (!do_tcpdrain)
911 return;
912
913 VNET_LIST_RLOCK_NOSLEEP();
914 VNET_FOREACH(vnet_iter) {
915 CURVNET_SET(vnet_iter);
916 struct inpcb *inpb;
917 struct tcpcb *tcpb;
918 struct tseg_qent *te;
919
920 /*
921 * Walk the tcpbs, if existing, and flush the reassembly queue,
922 * if there is one...
923 * XXX: The "Net/3" implementation doesn't imply that the TCP
924 * reassembly queue should be flushed, but in a situation
925 * where we're really low on mbufs, this is potentially
926 * usefull.
927 */
928 INP_INFO_RLOCK(&V_tcbinfo);
929 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
930 if (inpb->inp_flags & INP_TIMEWAIT)
931 continue;
932 INP_WLOCK(inpb);
933 if ((tcpb = intotcpcb(inpb)) != NULL) {
934 while ((te = LIST_FIRST(&tcpb->t_segq))
935 != NULL) {
936 LIST_REMOVE(te, tqe_q);
937 m_freem(te->tqe_m);
938 uma_zfree(V_tcp_reass_zone, te);
939 tcpb->t_segqlen--;
940 V_tcp_reass_qsize--;
941 }
942 tcp_clean_sackreport(tcpb);
943 }
944 INP_WUNLOCK(inpb);
945 }
946 INP_INFO_RUNLOCK(&V_tcbinfo);
947 CURVNET_RESTORE();
948 }
949 VNET_LIST_RUNLOCK_NOSLEEP();
950 }
951
952 /*
953 * Notify a tcp user of an asynchronous error;
954 * store error as soft error, but wake up user
955 * (for now, won't do anything until can select for soft error).
956 *
957 * Do not wake up user since there currently is no mechanism for
958 * reporting soft errors (yet - a kqueue filter may be added).
959 */
960 static struct inpcb *
961 tcp_notify(struct inpcb *inp, int error)
962 {
963 struct tcpcb *tp;
964
965 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
966 INP_WLOCK_ASSERT(inp);
967
968 if ((inp->inp_flags & INP_TIMEWAIT) ||
969 (inp->inp_flags & INP_DROPPED))
970 return (inp);
971
972 tp = intotcpcb(inp);
973 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
974
975 /*
976 * Ignore some errors if we are hooked up.
977 * If connection hasn't completed, has retransmitted several times,
978 * and receives a second error, give up now. This is better
979 * than waiting a long time to establish a connection that
980 * can never complete.
981 */
982 if (tp->t_state == TCPS_ESTABLISHED &&
983 (error == EHOSTUNREACH || error == ENETUNREACH ||
984 error == EHOSTDOWN)) {
985 return (inp);
986 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
987 tp->t_softerror) {
988 tp = tcp_drop(tp, error);
989 if (tp != NULL)
990 return (inp);
991 else
992 return (NULL);
993 } else {
994 tp->t_softerror = error;
995 return (inp);
996 }
997 #if 0
998 wakeup( &so->so_timeo);
999 sorwakeup(so);
1000 sowwakeup(so);
1001 #endif
1002 }
1003
1004 static int
1005 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1006 {
1007 int error, i, m, n, pcb_count;
1008 struct inpcb *inp, **inp_list;
1009 inp_gen_t gencnt;
1010 struct xinpgen xig;
1011
1012 /*
1013 * The process of preparing the TCB list is too time-consuming and
1014 * resource-intensive to repeat twice on every request.
1015 */
1016 if (req->oldptr == NULL) {
1017 m = syncache_pcbcount();
1018 n = V_tcbinfo.ipi_count;
1019 req->oldidx = 2 * (sizeof xig)
1020 + ((m + n) + n/8) * sizeof(struct xtcpcb);
1021 return (0);
1022 }
1023
1024 if (req->newptr != NULL)
1025 return (EPERM);
1026
1027 /*
1028 * OK, now we're committed to doing something.
1029 */
1030 INP_INFO_RLOCK(&V_tcbinfo);
1031 gencnt = V_tcbinfo.ipi_gencnt;
1032 n = V_tcbinfo.ipi_count;
1033 INP_INFO_RUNLOCK(&V_tcbinfo);
1034
1035 m = syncache_pcbcount();
1036
1037 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1038 + (n + m) * sizeof(struct xtcpcb));
1039 if (error != 0)
1040 return (error);
1041
1042 xig.xig_len = sizeof xig;
1043 xig.xig_count = n + m;
1044 xig.xig_gen = gencnt;
1045 xig.xig_sogen = so_gencnt;
1046 error = SYSCTL_OUT(req, &xig, sizeof xig);
1047 if (error)
1048 return (error);
1049
1050 error = syncache_pcblist(req, m, &pcb_count);
1051 if (error)
1052 return (error);
1053
1054 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1055 if (inp_list == NULL)
1056 return (ENOMEM);
1057
1058 INP_INFO_RLOCK(&V_tcbinfo);
1059 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1060 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1061 INP_WLOCK(inp);
1062 if (inp->inp_gencnt <= gencnt) {
1063 /*
1064 * XXX: This use of cr_cansee(), introduced with
1065 * TCP state changes, is not quite right, but for
1066 * now, better than nothing.
1067 */
1068 if (inp->inp_flags & INP_TIMEWAIT) {
1069 if (intotw(inp) != NULL)
1070 error = cr_cansee(req->td->td_ucred,
1071 intotw(inp)->tw_cred);
1072 else
1073 error = EINVAL; /* Skip this inp. */
1074 } else
1075 error = cr_canseeinpcb(req->td->td_ucred, inp);
1076 if (error == 0) {
1077 in_pcbref(inp);
1078 inp_list[i++] = inp;
1079 }
1080 }
1081 INP_WUNLOCK(inp);
1082 }
1083 INP_INFO_RUNLOCK(&V_tcbinfo);
1084 n = i;
1085
1086 error = 0;
1087 for (i = 0; i < n; i++) {
1088 inp = inp_list[i];
1089 INP_RLOCK(inp);
1090 if (inp->inp_gencnt <= gencnt) {
1091 struct xtcpcb xt;
1092 void *inp_ppcb;
1093
1094 bzero(&xt, sizeof(xt));
1095 xt.xt_len = sizeof xt;
1096 /* XXX should avoid extra copy */
1097 bcopy(inp, &xt.xt_inp, sizeof *inp);
1098 inp_ppcb = inp->inp_ppcb;
1099 if (inp_ppcb == NULL)
1100 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1101 else if (inp->inp_flags & INP_TIMEWAIT) {
1102 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1103 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1104 } else
1105 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1106 if (inp->inp_socket != NULL)
1107 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1108 else {
1109 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1110 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1111 }
1112 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1113 INP_RUNLOCK(inp);
1114 error = SYSCTL_OUT(req, &xt, sizeof xt);
1115 } else
1116 INP_RUNLOCK(inp);
1117 }
1118 INP_INFO_WLOCK(&V_tcbinfo);
1119 for (i = 0; i < n; i++) {
1120 inp = inp_list[i];
1121 INP_WLOCK(inp);
1122 if (!in_pcbrele(inp))
1123 INP_WUNLOCK(inp);
1124 }
1125 INP_INFO_WUNLOCK(&V_tcbinfo);
1126
1127 if (!error) {
1128 /*
1129 * Give the user an updated idea of our state.
1130 * If the generation differs from what we told
1131 * her before, she knows that something happened
1132 * while we were processing this request, and it
1133 * might be necessary to retry.
1134 */
1135 INP_INFO_RLOCK(&V_tcbinfo);
1136 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1137 xig.xig_sogen = so_gencnt;
1138 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1139 INP_INFO_RUNLOCK(&V_tcbinfo);
1140 error = SYSCTL_OUT(req, &xig, sizeof xig);
1141 }
1142 free(inp_list, M_TEMP);
1143 return (error);
1144 }
1145
1146 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1147 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1148
1149 static int
1150 tcp_getcred(SYSCTL_HANDLER_ARGS)
1151 {
1152 struct xucred xuc;
1153 struct sockaddr_in addrs[2];
1154 struct inpcb *inp;
1155 int error;
1156
1157 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1158 if (error)
1159 return (error);
1160 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1161 if (error)
1162 return (error);
1163 INP_INFO_RLOCK(&V_tcbinfo);
1164 inp = in_pcblookup_hash(&V_tcbinfo, addrs[1].sin_addr,
1165 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1166 if (inp != NULL) {
1167 INP_RLOCK(inp);
1168 INP_INFO_RUNLOCK(&V_tcbinfo);
1169 if (inp->inp_socket == NULL)
1170 error = ENOENT;
1171 if (error == 0)
1172 error = cr_canseeinpcb(req->td->td_ucred, inp);
1173 if (error == 0)
1174 cru2x(inp->inp_cred, &xuc);
1175 INP_RUNLOCK(inp);
1176 } else {
1177 INP_INFO_RUNLOCK(&V_tcbinfo);
1178 error = ENOENT;
1179 }
1180 if (error == 0)
1181 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1182 return (error);
1183 }
1184
1185 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1186 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1187 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1188
1189 #ifdef INET6
1190 static int
1191 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1192 {
1193 struct xucred xuc;
1194 struct sockaddr_in6 addrs[2];
1195 struct inpcb *inp;
1196 int error, mapped = 0;
1197
1198 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1199 if (error)
1200 return (error);
1201 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1202 if (error)
1203 return (error);
1204 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1205 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1206 return (error);
1207 }
1208 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1209 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1210 mapped = 1;
1211 else
1212 return (EINVAL);
1213 }
1214
1215 INP_INFO_RLOCK(&V_tcbinfo);
1216 if (mapped == 1)
1217 inp = in_pcblookup_hash(&V_tcbinfo,
1218 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1219 addrs[1].sin6_port,
1220 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1221 addrs[0].sin6_port,
1222 0, NULL);
1223 else
1224 inp = in6_pcblookup_hash(&V_tcbinfo,
1225 &addrs[1].sin6_addr, addrs[1].sin6_port,
1226 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1227 if (inp != NULL) {
1228 INP_RLOCK(inp);
1229 INP_INFO_RUNLOCK(&V_tcbinfo);
1230 if (inp->inp_socket == NULL)
1231 error = ENOENT;
1232 if (error == 0)
1233 error = cr_canseeinpcb(req->td->td_ucred, inp);
1234 if (error == 0)
1235 cru2x(inp->inp_cred, &xuc);
1236 INP_RUNLOCK(inp);
1237 } else {
1238 INP_INFO_RUNLOCK(&V_tcbinfo);
1239 error = ENOENT;
1240 }
1241 if (error == 0)
1242 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1243 return (error);
1244 }
1245
1246 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1247 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1248 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1249 #endif
1250
1251
1252 void
1253 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1254 {
1255 struct ip *ip = vip;
1256 struct tcphdr *th;
1257 struct in_addr faddr;
1258 struct inpcb *inp;
1259 struct tcpcb *tp;
1260 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1261 struct icmp *icp;
1262 struct in_conninfo inc;
1263 tcp_seq icmp_tcp_seq;
1264 int mtu;
1265
1266 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1267 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1268 return;
1269
1270 if (cmd == PRC_MSGSIZE)
1271 notify = tcp_mtudisc;
1272 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1273 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1274 notify = tcp_drop_syn_sent;
1275 /*
1276 * Redirects don't need to be handled up here.
1277 */
1278 else if (PRC_IS_REDIRECT(cmd))
1279 return;
1280 /*
1281 * Source quench is depreciated.
1282 */
1283 else if (cmd == PRC_QUENCH)
1284 return;
1285 /*
1286 * Hostdead is ugly because it goes linearly through all PCBs.
1287 * XXX: We never get this from ICMP, otherwise it makes an
1288 * excellent DoS attack on machines with many connections.
1289 */
1290 else if (cmd == PRC_HOSTDEAD)
1291 ip = NULL;
1292 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1293 return;
1294 if (ip != NULL) {
1295 icp = (struct icmp *)((caddr_t)ip
1296 - offsetof(struct icmp, icmp_ip));
1297 th = (struct tcphdr *)((caddr_t)ip
1298 + (ip->ip_hl << 2));
1299 INP_INFO_WLOCK(&V_tcbinfo);
1300 inp = in_pcblookup_hash(&V_tcbinfo, faddr, th->th_dport,
1301 ip->ip_src, th->th_sport, 0, NULL);
1302 if (inp != NULL) {
1303 INP_WLOCK(inp);
1304 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1305 !(inp->inp_flags & INP_DROPPED) &&
1306 !(inp->inp_socket == NULL)) {
1307 icmp_tcp_seq = htonl(th->th_seq);
1308 tp = intotcpcb(inp);
1309 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1310 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1311 if (cmd == PRC_MSGSIZE) {
1312 /*
1313 * MTU discovery:
1314 * If we got a needfrag set the MTU
1315 * in the route to the suggested new
1316 * value (if given) and then notify.
1317 */
1318 bzero(&inc, sizeof(inc));
1319 inc.inc_faddr = faddr;
1320 inc.inc_fibnum =
1321 inp->inp_inc.inc_fibnum;
1322
1323 mtu = ntohs(icp->icmp_nextmtu);
1324 /*
1325 * If no alternative MTU was
1326 * proposed, try the next smaller
1327 * one. ip->ip_len has already
1328 * been swapped in icmp_input().
1329 */
1330 if (!mtu)
1331 mtu = ip_next_mtu(ip->ip_len,
1332 1);
1333 if (mtu < max(296, V_tcp_minmss
1334 + sizeof(struct tcpiphdr)))
1335 mtu = 0;
1336 if (!mtu)
1337 mtu = V_tcp_mssdflt
1338 + sizeof(struct tcpiphdr);
1339 /*
1340 * Only cache the the MTU if it
1341 * is smaller than the interface
1342 * or route MTU. tcp_mtudisc()
1343 * will do right thing by itself.
1344 */
1345 if (mtu <= tcp_maxmtu(&inc, NULL))
1346 tcp_hc_updatemtu(&inc, mtu);
1347 }
1348
1349 inp = (*notify)(inp, inetctlerrmap[cmd]);
1350 }
1351 }
1352 if (inp != NULL)
1353 INP_WUNLOCK(inp);
1354 } else {
1355 bzero(&inc, sizeof(inc));
1356 inc.inc_fport = th->th_dport;
1357 inc.inc_lport = th->th_sport;
1358 inc.inc_faddr = faddr;
1359 inc.inc_laddr = ip->ip_src;
1360 syncache_unreach(&inc, th);
1361 }
1362 INP_INFO_WUNLOCK(&V_tcbinfo);
1363 } else
1364 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1365 }
1366
1367 #ifdef INET6
1368 void
1369 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1370 {
1371 struct tcphdr th;
1372 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1373 struct ip6_hdr *ip6;
1374 struct mbuf *m;
1375 struct ip6ctlparam *ip6cp = NULL;
1376 const struct sockaddr_in6 *sa6_src = NULL;
1377 int off;
1378 struct tcp_portonly {
1379 u_int16_t th_sport;
1380 u_int16_t th_dport;
1381 } *thp;
1382
1383 if (sa->sa_family != AF_INET6 ||
1384 sa->sa_len != sizeof(struct sockaddr_in6))
1385 return;
1386
1387 if (cmd == PRC_MSGSIZE)
1388 notify = tcp_mtudisc;
1389 else if (!PRC_IS_REDIRECT(cmd) &&
1390 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1391 return;
1392 /* Source quench is depreciated. */
1393 else if (cmd == PRC_QUENCH)
1394 return;
1395
1396 /* if the parameter is from icmp6, decode it. */
1397 if (d != NULL) {
1398 ip6cp = (struct ip6ctlparam *)d;
1399 m = ip6cp->ip6c_m;
1400 ip6 = ip6cp->ip6c_ip6;
1401 off = ip6cp->ip6c_off;
1402 sa6_src = ip6cp->ip6c_src;
1403 } else {
1404 m = NULL;
1405 ip6 = NULL;
1406 off = 0; /* fool gcc */
1407 sa6_src = &sa6_any;
1408 }
1409
1410 if (ip6 != NULL) {
1411 struct in_conninfo inc;
1412 /*
1413 * XXX: We assume that when IPV6 is non NULL,
1414 * M and OFF are valid.
1415 */
1416
1417 /* check if we can safely examine src and dst ports */
1418 if (m->m_pkthdr.len < off + sizeof(*thp))
1419 return;
1420
1421 bzero(&th, sizeof(th));
1422 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1423
1424 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1425 (struct sockaddr *)ip6cp->ip6c_src,
1426 th.th_sport, cmd, NULL, notify);
1427
1428 bzero(&inc, sizeof(inc));
1429 inc.inc_fport = th.th_dport;
1430 inc.inc_lport = th.th_sport;
1431 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1432 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1433 inc.inc_flags |= INC_ISIPV6;
1434 INP_INFO_WLOCK(&V_tcbinfo);
1435 syncache_unreach(&inc, &th);
1436 INP_INFO_WUNLOCK(&V_tcbinfo);
1437 } else
1438 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1439 0, cmd, NULL, notify);
1440 }
1441 #endif /* INET6 */
1442
1443
1444 /*
1445 * Following is where TCP initial sequence number generation occurs.
1446 *
1447 * There are two places where we must use initial sequence numbers:
1448 * 1. In SYN-ACK packets.
1449 * 2. In SYN packets.
1450 *
1451 * All ISNs for SYN-ACK packets are generated by the syncache. See
1452 * tcp_syncache.c for details.
1453 *
1454 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1455 * depends on this property. In addition, these ISNs should be
1456 * unguessable so as to prevent connection hijacking. To satisfy
1457 * the requirements of this situation, the algorithm outlined in
1458 * RFC 1948 is used, with only small modifications.
1459 *
1460 * Implementation details:
1461 *
1462 * Time is based off the system timer, and is corrected so that it
1463 * increases by one megabyte per second. This allows for proper
1464 * recycling on high speed LANs while still leaving over an hour
1465 * before rollover.
1466 *
1467 * As reading the *exact* system time is too expensive to be done
1468 * whenever setting up a TCP connection, we increment the time
1469 * offset in two ways. First, a small random positive increment
1470 * is added to isn_offset for each connection that is set up.
1471 * Second, the function tcp_isn_tick fires once per clock tick
1472 * and increments isn_offset as necessary so that sequence numbers
1473 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1474 * random positive increments serve only to ensure that the same
1475 * exact sequence number is never sent out twice (as could otherwise
1476 * happen when a port is recycled in less than the system tick
1477 * interval.)
1478 *
1479 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1480 * between seeding of isn_secret. This is normally set to zero,
1481 * as reseeding should not be necessary.
1482 *
1483 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1484 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1485 * general, this means holding an exclusive (write) lock.
1486 */
1487
1488 #define ISN_BYTES_PER_SECOND 1048576
1489 #define ISN_STATIC_INCREMENT 4096
1490 #define ISN_RANDOM_INCREMENT (4096 - 1)
1491
1492 static VNET_DEFINE(u_char, isn_secret[32]);
1493 static VNET_DEFINE(int, isn_last_reseed);
1494 static VNET_DEFINE(u_int32_t, isn_offset);
1495 static VNET_DEFINE(u_int32_t, isn_offset_old);
1496
1497 #define V_isn_secret VNET(isn_secret)
1498 #define V_isn_last_reseed VNET(isn_last_reseed)
1499 #define V_isn_offset VNET(isn_offset)
1500 #define V_isn_offset_old VNET(isn_offset_old)
1501
1502 tcp_seq
1503 tcp_new_isn(struct tcpcb *tp)
1504 {
1505 MD5_CTX isn_ctx;
1506 u_int32_t md5_buffer[4];
1507 tcp_seq new_isn;
1508
1509 INP_WLOCK_ASSERT(tp->t_inpcb);
1510
1511 ISN_LOCK();
1512 /* Seed if this is the first use, reseed if requested. */
1513 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1514 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1515 < (u_int)ticks))) {
1516 read_random(&V_isn_secret, sizeof(V_isn_secret));
1517 V_isn_last_reseed = ticks;
1518 }
1519
1520 /* Compute the md5 hash and return the ISN. */
1521 MD5Init(&isn_ctx);
1522 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1523 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1524 #ifdef INET6
1525 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1526 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1527 sizeof(struct in6_addr));
1528 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1529 sizeof(struct in6_addr));
1530 } else
1531 #endif
1532 {
1533 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1534 sizeof(struct in_addr));
1535 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1536 sizeof(struct in_addr));
1537 }
1538 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1539 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1540 new_isn = (tcp_seq) md5_buffer[0];
1541 V_isn_offset += ISN_STATIC_INCREMENT +
1542 (arc4random() & ISN_RANDOM_INCREMENT);
1543 new_isn += V_isn_offset;
1544 ISN_UNLOCK();
1545 return (new_isn);
1546 }
1547
1548 /*
1549 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1550 * to keep time flowing at a relatively constant rate. If the random
1551 * increments have already pushed us past the projected offset, do nothing.
1552 */
1553 static void
1554 tcp_isn_tick(void *xtp)
1555 {
1556 VNET_ITERATOR_DECL(vnet_iter);
1557 u_int32_t projected_offset;
1558
1559 VNET_LIST_RLOCK_NOSLEEP();
1560 ISN_LOCK();
1561 VNET_FOREACH(vnet_iter) {
1562 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS */
1563 projected_offset =
1564 V_isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1565
1566 if (SEQ_GT(projected_offset, V_isn_offset))
1567 V_isn_offset = projected_offset;
1568
1569 V_isn_offset_old = V_isn_offset;
1570 CURVNET_RESTORE();
1571 }
1572 ISN_UNLOCK();
1573 VNET_LIST_RUNLOCK_NOSLEEP();
1574 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1575 }
1576
1577 /*
1578 * When a specific ICMP unreachable message is received and the
1579 * connection state is SYN-SENT, drop the connection. This behavior
1580 * is controlled by the icmp_may_rst sysctl.
1581 */
1582 struct inpcb *
1583 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1584 {
1585 struct tcpcb *tp;
1586
1587 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1588 INP_WLOCK_ASSERT(inp);
1589
1590 if ((inp->inp_flags & INP_TIMEWAIT) ||
1591 (inp->inp_flags & INP_DROPPED))
1592 return (inp);
1593
1594 tp = intotcpcb(inp);
1595 if (tp->t_state != TCPS_SYN_SENT)
1596 return (inp);
1597
1598 tp = tcp_drop(tp, errno);
1599 if (tp != NULL)
1600 return (inp);
1601 else
1602 return (NULL);
1603 }
1604
1605 /*
1606 * When `need fragmentation' ICMP is received, update our idea of the MSS
1607 * based on the new value in the route. Also nudge TCP to send something,
1608 * since we know the packet we just sent was dropped.
1609 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1610 */
1611 struct inpcb *
1612 tcp_mtudisc(struct inpcb *inp, int errno)
1613 {
1614 struct tcpcb *tp;
1615 struct socket *so;
1616
1617 INP_WLOCK_ASSERT(inp);
1618 if ((inp->inp_flags & INP_TIMEWAIT) ||
1619 (inp->inp_flags & INP_DROPPED))
1620 return (inp);
1621
1622 tp = intotcpcb(inp);
1623 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1624
1625 tcp_mss_update(tp, -1, NULL, NULL);
1626
1627 so = inp->inp_socket;
1628 SOCKBUF_LOCK(&so->so_snd);
1629 /* If the mss is larger than the socket buffer, decrease the mss. */
1630 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1631 tp->t_maxseg = so->so_snd.sb_hiwat;
1632 SOCKBUF_UNLOCK(&so->so_snd);
1633
1634 TCPSTAT_INC(tcps_mturesent);
1635 tp->t_rtttime = 0;
1636 tp->snd_nxt = tp->snd_una;
1637 tcp_free_sackholes(tp);
1638 tp->snd_recover = tp->snd_max;
1639 if (tp->t_flags & TF_SACK_PERMIT)
1640 EXIT_FASTRECOVERY(tp);
1641 tcp_output_send(tp);
1642 return (inp);
1643 }
1644
1645 /*
1646 * Look-up the routing entry to the peer of this inpcb. If no route
1647 * is found and it cannot be allocated, then return 0. This routine
1648 * is called by TCP routines that access the rmx structure and by
1649 * tcp_mss_update to get the peer/interface MTU.
1650 */
1651 u_long
1652 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1653 {
1654 struct route sro;
1655 struct sockaddr_in *dst;
1656 struct ifnet *ifp;
1657 u_long maxmtu = 0;
1658
1659 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1660
1661 bzero(&sro, sizeof(sro));
1662 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1663 dst = (struct sockaddr_in *)&sro.ro_dst;
1664 dst->sin_family = AF_INET;
1665 dst->sin_len = sizeof(*dst);
1666 dst->sin_addr = inc->inc_faddr;
1667 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1668 }
1669 if (sro.ro_rt != NULL) {
1670 ifp = sro.ro_rt->rt_ifp;
1671 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1672 maxmtu = ifp->if_mtu;
1673 else
1674 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1675
1676 /* Report additional interface capabilities. */
1677 if (flags != NULL) {
1678 if (ifp->if_capenable & IFCAP_TSO4 &&
1679 ifp->if_hwassist & CSUM_TSO)
1680 *flags |= CSUM_TSO;
1681 }
1682 RTFREE(sro.ro_rt);
1683 }
1684 return (maxmtu);
1685 }
1686
1687 #ifdef INET6
1688 u_long
1689 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1690 {
1691 struct route_in6 sro6;
1692 struct ifnet *ifp;
1693 u_long maxmtu = 0;
1694
1695 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1696
1697 bzero(&sro6, sizeof(sro6));
1698 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1699 sro6.ro_dst.sin6_family = AF_INET6;
1700 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1701 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1702 rtalloc_ign((struct route *)&sro6, 0);
1703 }
1704 if (sro6.ro_rt != NULL) {
1705 ifp = sro6.ro_rt->rt_ifp;
1706 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1707 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1708 else
1709 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1710 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1711
1712 /* Report additional interface capabilities. */
1713 if (flags != NULL) {
1714 if (ifp->if_capenable & IFCAP_TSO6 &&
1715 ifp->if_hwassist & CSUM_TSO)
1716 *flags |= CSUM_TSO;
1717 }
1718 RTFREE(sro6.ro_rt);
1719 }
1720
1721 return (maxmtu);
1722 }
1723 #endif /* INET6 */
1724
1725 #ifdef IPSEC
1726 /* compute ESP/AH header size for TCP, including outer IP header. */
1727 size_t
1728 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1729 {
1730 struct inpcb *inp;
1731 struct mbuf *m;
1732 size_t hdrsiz;
1733 struct ip *ip;
1734 #ifdef INET6
1735 struct ip6_hdr *ip6;
1736 #endif
1737 struct tcphdr *th;
1738
1739 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1740 return (0);
1741 MGETHDR(m, M_DONTWAIT, MT_DATA);
1742 if (!m)
1743 return (0);
1744
1745 #ifdef INET6
1746 if ((inp->inp_vflag & INP_IPV6) != 0) {
1747 ip6 = mtod(m, struct ip6_hdr *);
1748 th = (struct tcphdr *)(ip6 + 1);
1749 m->m_pkthdr.len = m->m_len =
1750 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1751 tcpip_fillheaders(inp, ip6, th);
1752 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1753 } else
1754 #endif /* INET6 */
1755 {
1756 ip = mtod(m, struct ip *);
1757 th = (struct tcphdr *)(ip + 1);
1758 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1759 tcpip_fillheaders(inp, ip, th);
1760 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1761 }
1762
1763 m_free(m);
1764 return (hdrsiz);
1765 }
1766 #endif /* IPSEC */
1767
1768 /*
1769 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1770 *
1771 * This code attempts to calculate the bandwidth-delay product as a
1772 * means of determining the optimal window size to maximize bandwidth,
1773 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1774 * routers. This code also does a fairly good job keeping RTTs in check
1775 * across slow links like modems. We implement an algorithm which is very
1776 * similar (but not meant to be) TCP/Vegas. The code operates on the
1777 * transmitter side of a TCP connection and so only effects the transmit
1778 * side of the connection.
1779 *
1780 * BACKGROUND: TCP makes no provision for the management of buffer space
1781 * at the end points or at the intermediate routers and switches. A TCP
1782 * stream, whether using NewReno or not, will eventually buffer as
1783 * many packets as it is able and the only reason this typically works is
1784 * due to the fairly small default buffers made available for a connection
1785 * (typicaly 16K or 32K). As machines use larger windows and/or window
1786 * scaling it is now fairly easy for even a single TCP connection to blow-out
1787 * all available buffer space not only on the local interface, but on
1788 * intermediate routers and switches as well. NewReno makes a misguided
1789 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1790 * then backing off, then steadily increasing the window again until another
1791 * failure occurs, ad-infinitum. This results in terrible oscillation that
1792 * is only made worse as network loads increase and the idea of intentionally
1793 * blowing out network buffers is, frankly, a terrible way to manage network
1794 * resources.
1795 *
1796 * It is far better to limit the transmit window prior to the failure
1797 * condition being achieved. There are two general ways to do this: First
1798 * you can 'scan' through different transmit window sizes and locate the
1799 * point where the RTT stops increasing, indicating that you have filled the
1800 * pipe, then scan backwards until you note that RTT stops decreasing, then
1801 * repeat ad-infinitum. This method works in principle but has severe
1802 * implementation issues due to RTT variances, timer granularity, and
1803 * instability in the algorithm which can lead to many false positives and
1804 * create oscillations as well as interact badly with other TCP streams
1805 * implementing the same algorithm.
1806 *
1807 * The second method is to limit the window to the bandwidth delay product
1808 * of the link. This is the method we implement. RTT variances and our
1809 * own manipulation of the congestion window, bwnd, can potentially
1810 * destabilize the algorithm. For this reason we have to stabilize the
1811 * elements used to calculate the window. We do this by using the minimum
1812 * observed RTT, the long term average of the observed bandwidth, and
1813 * by adding two segments worth of slop. It isn't perfect but it is able
1814 * to react to changing conditions and gives us a very stable basis on
1815 * which to extend the algorithm.
1816 */
1817 void
1818 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1819 {
1820 u_long bw;
1821 u_long bwnd;
1822 int save_ticks;
1823
1824 INP_WLOCK_ASSERT(tp->t_inpcb);
1825
1826 /*
1827 * If inflight_enable is disabled in the middle of a tcp connection,
1828 * make sure snd_bwnd is effectively disabled.
1829 */
1830 if (V_tcp_inflight_enable == 0 ||
1831 tp->t_rttlow < V_tcp_inflight_rttthresh) {
1832 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1833 tp->snd_bandwidth = 0;
1834 return;
1835 }
1836
1837 /*
1838 * Figure out the bandwidth. Due to the tick granularity this
1839 * is a very rough number and it MUST be averaged over a fairly
1840 * long period of time. XXX we need to take into account a link
1841 * that is not using all available bandwidth, but for now our
1842 * slop will ramp us up if this case occurs and the bandwidth later
1843 * increases.
1844 *
1845 * Note: if ticks rollover 'bw' may wind up negative. We must
1846 * effectively reset t_bw_rtttime for this case.
1847 */
1848 save_ticks = ticks;
1849 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1850 return;
1851
1852 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1853 (save_ticks - tp->t_bw_rtttime);
1854 tp->t_bw_rtttime = save_ticks;
1855 tp->t_bw_rtseq = ack_seq;
1856 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1857 return;
1858 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1859
1860 tp->snd_bandwidth = bw;
1861
1862 /*
1863 * Calculate the semi-static bandwidth delay product, plus two maximal
1864 * segments. The additional slop puts us squarely in the sweet
1865 * spot and also handles the bandwidth run-up case and stabilization.
1866 * Without the slop we could be locking ourselves into a lower
1867 * bandwidth.
1868 *
1869 * Situations Handled:
1870 * (1) Prevents over-queueing of packets on LANs, especially on
1871 * high speed LANs, allowing larger TCP buffers to be
1872 * specified, and also does a good job preventing
1873 * over-queueing of packets over choke points like modems
1874 * (at least for the transmit side).
1875 *
1876 * (2) Is able to handle changing network loads (bandwidth
1877 * drops so bwnd drops, bandwidth increases so bwnd
1878 * increases).
1879 *
1880 * (3) Theoretically should stabilize in the face of multiple
1881 * connections implementing the same algorithm (this may need
1882 * a little work).
1883 *
1884 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1885 * be adjusted with a sysctl but typically only needs to be
1886 * on very slow connections. A value no smaller then 5
1887 * should be used, but only reduce this default if you have
1888 * no other choice.
1889 */
1890 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1891 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + V_tcp_inflight_stab * tp->t_maxseg / 10;
1892 #undef USERTT
1893
1894 if (tcp_inflight_debug > 0) {
1895 static int ltime;
1896 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1897 ltime = ticks;
1898 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1899 tp,
1900 bw,
1901 tp->t_rttbest,
1902 tp->t_srtt,
1903 bwnd
1904 );
1905 }
1906 }
1907 if ((long)bwnd < V_tcp_inflight_min)
1908 bwnd = V_tcp_inflight_min;
1909 if (bwnd > V_tcp_inflight_max)
1910 bwnd = V_tcp_inflight_max;
1911 if ((long)bwnd < tp->t_maxseg * 2)
1912 bwnd = tp->t_maxseg * 2;
1913 tp->snd_bwnd = bwnd;
1914 }
1915
1916 #ifdef TCP_SIGNATURE
1917 /*
1918 * Callback function invoked by m_apply() to digest TCP segment data
1919 * contained within an mbuf chain.
1920 */
1921 static int
1922 tcp_signature_apply(void *fstate, void *data, u_int len)
1923 {
1924
1925 MD5Update(fstate, (u_char *)data, len);
1926 return (0);
1927 }
1928
1929 /*
1930 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1931 *
1932 * Parameters:
1933 * m pointer to head of mbuf chain
1934 * _unused
1935 * len length of TCP segment data, excluding options
1936 * optlen length of TCP segment options
1937 * buf pointer to storage for computed MD5 digest
1938 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1939 *
1940 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1941 * When called from tcp_input(), we can be sure that th_sum has been
1942 * zeroed out and verified already.
1943 *
1944 * Return 0 if successful, otherwise return -1.
1945 *
1946 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1947 * search with the destination IP address, and a 'magic SPI' to be
1948 * determined by the application. This is hardcoded elsewhere to 1179
1949 * right now. Another branch of this code exists which uses the SPD to
1950 * specify per-application flows but it is unstable.
1951 */
1952 int
1953 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1954 u_char *buf, u_int direction)
1955 {
1956 union sockaddr_union dst;
1957 struct ippseudo ippseudo;
1958 MD5_CTX ctx;
1959 int doff;
1960 struct ip *ip;
1961 struct ipovly *ipovly;
1962 struct secasvar *sav;
1963 struct tcphdr *th;
1964 #ifdef INET6
1965 struct ip6_hdr *ip6;
1966 struct in6_addr in6;
1967 char ip6buf[INET6_ADDRSTRLEN];
1968 uint32_t plen;
1969 uint16_t nhdr;
1970 #endif
1971 u_short savecsum;
1972
1973 KASSERT(m != NULL, ("NULL mbuf chain"));
1974 KASSERT(buf != NULL, ("NULL signature pointer"));
1975
1976 /* Extract the destination from the IP header in the mbuf. */
1977 bzero(&dst, sizeof(union sockaddr_union));
1978 ip = mtod(m, struct ip *);
1979 #ifdef INET6
1980 ip6 = NULL; /* Make the compiler happy. */
1981 #endif
1982 switch (ip->ip_v) {
1983 case IPVERSION:
1984 dst.sa.sa_len = sizeof(struct sockaddr_in);
1985 dst.sa.sa_family = AF_INET;
1986 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1987 ip->ip_src : ip->ip_dst;
1988 break;
1989 #ifdef INET6
1990 case (IPV6_VERSION >> 4):
1991 ip6 = mtod(m, struct ip6_hdr *);
1992 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1993 dst.sa.sa_family = AF_INET6;
1994 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1995 ip6->ip6_src : ip6->ip6_dst;
1996 break;
1997 #endif
1998 default:
1999 return (EINVAL);
2000 /* NOTREACHED */
2001 break;
2002 }
2003
2004 /* Look up an SADB entry which matches the address of the peer. */
2005 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2006 if (sav == NULL) {
2007 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2008 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2009 #ifdef INET6
2010 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2011 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2012 #endif
2013 "(unsupported)"));
2014 return (EINVAL);
2015 }
2016
2017 MD5Init(&ctx);
2018 /*
2019 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2020 *
2021 * XXX The ippseudo header MUST be digested in network byte order,
2022 * or else we'll fail the regression test. Assume all fields we've
2023 * been doing arithmetic on have been in host byte order.
2024 * XXX One cannot depend on ipovly->ih_len here. When called from
2025 * tcp_output(), the underlying ip_len member has not yet been set.
2026 */
2027 switch (ip->ip_v) {
2028 case IPVERSION:
2029 ipovly = (struct ipovly *)ip;
2030 ippseudo.ippseudo_src = ipovly->ih_src;
2031 ippseudo.ippseudo_dst = ipovly->ih_dst;
2032 ippseudo.ippseudo_pad = 0;
2033 ippseudo.ippseudo_p = IPPROTO_TCP;
2034 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2035 optlen);
2036 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2037
2038 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2039 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2040 break;
2041 #ifdef INET6
2042 /*
2043 * RFC 2385, 2.0 Proposal
2044 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2045 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2046 * extended next header value (to form 32 bits), and 32-bit segment
2047 * length.
2048 * Note: Upper-Layer Packet Length comes before Next Header.
2049 */
2050 case (IPV6_VERSION >> 4):
2051 in6 = ip6->ip6_src;
2052 in6_clearscope(&in6);
2053 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2054 in6 = ip6->ip6_dst;
2055 in6_clearscope(&in6);
2056 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2057 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2058 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2059 nhdr = 0;
2060 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2061 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2062 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2063 nhdr = IPPROTO_TCP;
2064 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2065
2066 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2067 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2068 break;
2069 #endif
2070 default:
2071 return (EINVAL);
2072 /* NOTREACHED */
2073 break;
2074 }
2075
2076
2077 /*
2078 * Step 2: Update MD5 hash with TCP header, excluding options.
2079 * The TCP checksum must be set to zero.
2080 */
2081 savecsum = th->th_sum;
2082 th->th_sum = 0;
2083 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2084 th->th_sum = savecsum;
2085
2086 /*
2087 * Step 3: Update MD5 hash with TCP segment data.
2088 * Use m_apply() to avoid an early m_pullup().
2089 */
2090 if (len > 0)
2091 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2092
2093 /*
2094 * Step 4: Update MD5 hash with shared secret.
2095 */
2096 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2097 MD5Final(buf, &ctx);
2098
2099 key_sa_recordxfer(sav, m);
2100 KEY_FREESAV(&sav);
2101 return (0);
2102 }
2103 #endif /* TCP_SIGNATURE */
2104
2105 static int
2106 sysctl_drop(SYSCTL_HANDLER_ARGS)
2107 {
2108 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2109 struct sockaddr_storage addrs[2];
2110 struct inpcb *inp;
2111 struct tcpcb *tp;
2112 struct tcptw *tw;
2113 struct sockaddr_in *fin, *lin;
2114 #ifdef INET6
2115 struct sockaddr_in6 *fin6, *lin6;
2116 #endif
2117 int error;
2118
2119 inp = NULL;
2120 fin = lin = NULL;
2121 #ifdef INET6
2122 fin6 = lin6 = NULL;
2123 #endif
2124 error = 0;
2125
2126 if (req->oldptr != NULL || req->oldlen != 0)
2127 return (EINVAL);
2128 if (req->newptr == NULL)
2129 return (EPERM);
2130 if (req->newlen < sizeof(addrs))
2131 return (ENOMEM);
2132 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2133 if (error)
2134 return (error);
2135
2136 switch (addrs[0].ss_family) {
2137 #ifdef INET6
2138 case AF_INET6:
2139 fin6 = (struct sockaddr_in6 *)&addrs[0];
2140 lin6 = (struct sockaddr_in6 *)&addrs[1];
2141 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2142 lin6->sin6_len != sizeof(struct sockaddr_in6))
2143 return (EINVAL);
2144 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2145 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2146 return (EINVAL);
2147 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2148 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2149 fin = (struct sockaddr_in *)&addrs[0];
2150 lin = (struct sockaddr_in *)&addrs[1];
2151 break;
2152 }
2153 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2154 if (error)
2155 return (error);
2156 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2157 if (error)
2158 return (error);
2159 break;
2160 #endif
2161 case AF_INET:
2162 fin = (struct sockaddr_in *)&addrs[0];
2163 lin = (struct sockaddr_in *)&addrs[1];
2164 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2165 lin->sin_len != sizeof(struct sockaddr_in))
2166 return (EINVAL);
2167 break;
2168 default:
2169 return (EINVAL);
2170 }
2171 INP_INFO_WLOCK(&V_tcbinfo);
2172 switch (addrs[0].ss_family) {
2173 #ifdef INET6
2174 case AF_INET6:
2175 inp = in6_pcblookup_hash(&V_tcbinfo, &fin6->sin6_addr,
2176 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 0,
2177 NULL);
2178 break;
2179 #endif
2180 case AF_INET:
2181 inp = in_pcblookup_hash(&V_tcbinfo, fin->sin_addr,
2182 fin->sin_port, lin->sin_addr, lin->sin_port, 0, NULL);
2183 break;
2184 }
2185 if (inp != NULL) {
2186 INP_WLOCK(inp);
2187 if (inp->inp_flags & INP_TIMEWAIT) {
2188 /*
2189 * XXXRW: There currently exists a state where an
2190 * inpcb is present, but its timewait state has been
2191 * discarded. For now, don't allow dropping of this
2192 * type of inpcb.
2193 */
2194 tw = intotw(inp);
2195 if (tw != NULL)
2196 tcp_twclose(tw, 0);
2197 else
2198 INP_WUNLOCK(inp);
2199 } else if (!(inp->inp_flags & INP_DROPPED) &&
2200 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2201 tp = intotcpcb(inp);
2202 tp = tcp_drop(tp, ECONNABORTED);
2203 if (tp != NULL)
2204 INP_WUNLOCK(inp);
2205 } else
2206 INP_WUNLOCK(inp);
2207 } else
2208 error = ESRCH;
2209 INP_INFO_WUNLOCK(&V_tcbinfo);
2210 return (error);
2211 }
2212
2213 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2214 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2215 0, sysctl_drop, "", "Drop TCP connection");
2216
2217 /*
2218 * Generate a standardized TCP log line for use throughout the
2219 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2220 * allow use in the interrupt context.
2221 *
2222 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2223 * NB: The function may return NULL if memory allocation failed.
2224 *
2225 * Due to header inclusion and ordering limitations the struct ip
2226 * and ip6_hdr pointers have to be passed as void pointers.
2227 */
2228 char *
2229 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2230 const void *ip6hdr)
2231 {
2232 char *s, *sp;
2233 size_t size;
2234 struct ip *ip;
2235 #ifdef INET6
2236 const struct ip6_hdr *ip6;
2237
2238 ip6 = (const struct ip6_hdr *)ip6hdr;
2239 #endif /* INET6 */
2240 ip = (struct ip *)ip4hdr;
2241
2242 /*
2243 * The log line looks like this:
2244 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2245 */
2246 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2247 sizeof(PRINT_TH_FLAGS) + 1 +
2248 #ifdef INET6
2249 2 * INET6_ADDRSTRLEN;
2250 #else
2251 2 * INET_ADDRSTRLEN;
2252 #endif /* INET6 */
2253
2254 /* Is logging enabled? */
2255 if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2256 return (NULL);
2257
2258 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2259 if (s == NULL)
2260 return (NULL);
2261
2262 strcat(s, "TCP: [");
2263 sp = s + strlen(s);
2264
2265 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2266 inet_ntoa_r(inc->inc_faddr, sp);
2267 sp = s + strlen(s);
2268 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2269 sp = s + strlen(s);
2270 inet_ntoa_r(inc->inc_laddr, sp);
2271 sp = s + strlen(s);
2272 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2273 #ifdef INET6
2274 } else if (inc) {
2275 ip6_sprintf(sp, &inc->inc6_faddr);
2276 sp = s + strlen(s);
2277 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2278 sp = s + strlen(s);
2279 ip6_sprintf(sp, &inc->inc6_laddr);
2280 sp = s + strlen(s);
2281 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2282 } else if (ip6 && th) {
2283 ip6_sprintf(sp, &ip6->ip6_src);
2284 sp = s + strlen(s);
2285 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2286 sp = s + strlen(s);
2287 ip6_sprintf(sp, &ip6->ip6_dst);
2288 sp = s + strlen(s);
2289 sprintf(sp, "]:%i", ntohs(th->th_dport));
2290 #endif /* INET6 */
2291 } else if (ip && th) {
2292 inet_ntoa_r(ip->ip_src, sp);
2293 sp = s + strlen(s);
2294 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2295 sp = s + strlen(s);
2296 inet_ntoa_r(ip->ip_dst, sp);
2297 sp = s + strlen(s);
2298 sprintf(sp, "]:%i", ntohs(th->th_dport));
2299 } else {
2300 free(s, M_TCPLOG);
2301 return (NULL);
2302 }
2303 sp = s + strlen(s);
2304 if (th)
2305 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2306 if (*(s + size - 1) != '\0')
2307 panic("%s: string too long", __func__);
2308 return (s);
2309 }
Cache object: b5a2fa9ac1828e7572b911c04ab2b02f
|